blob: 0a1f8312b4dcf0fe9272310f1ab6d00178cdd2a8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/file.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * Changes Copyright (C) 1994 by Florian La Roche
7 * - Do not copy data too often around in the kernel.
8 * - In nfs_file_read the return value of kmalloc wasn't checked.
9 * - Put in a better version of read look-ahead buffering. Original idea
10 * and implementation by Wai S Kok elekokws@ee.nus.sg.
11 *
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
13 *
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
15 *
16 * nfs regular file handling functions
17 */
18
19#include <linux/time.h>
20#include <linux/kernel.h>
21#include <linux/errno.h>
22#include <linux/fcntl.h>
23#include <linux/stat.h>
24#include <linux/nfs_fs.h>
25#include <linux/nfs_mount.h>
26#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/pagemap.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040028#include <linux/aio.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090029#include <linux/gfp.h>
Trond Myklebustb608b282010-07-30 15:31:54 -040030#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/uaccess.h>
33#include <asm/system.h>
34
35#include "delegation.h"
Trond Myklebust94387fb2007-07-22 17:09:05 -040036#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050037#include "iostat.h"
David Howells545db452009-04-03 16:42:44 +010038#include "fscache.h"
Benny Halevye5e94012010-10-20 00:18:01 -040039#include "pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#define NFSDBG_FACILITY NFSDBG_FILE
42
43static int nfs_file_open(struct inode *, struct file *);
44static int nfs_file_release(struct inode *, struct file *);
Trond Myklebust980802e2005-06-13 11:14:01 -040045static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046static int nfs_file_mmap(struct file *, struct vm_area_struct *);
Jens Axboef0930ff2007-06-01 11:51:43 +020047static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
48 struct pipe_inode_info *pipe,
49 size_t count, unsigned int flags);
Badari Pulavarty027445c2006-09-30 23:28:46 -070050static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
51 unsigned long nr_segs, loff_t pos);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -070052static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
53 struct file *filp, loff_t *ppos,
54 size_t count, unsigned int flags);
Badari Pulavarty027445c2006-09-30 23:28:46 -070055static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
56 unsigned long nr_segs, loff_t pos);
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -070057static int nfs_file_flush(struct file *, fl_owner_t id);
Josef Bacik02c24a82011-07-16 20:44:56 -040058static int nfs_file_fsync(struct file *, loff_t, loff_t, int datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059static int nfs_check_flags(int flags);
60static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
61static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
J. Bruce Fields370f6592007-06-08 15:23:34 -040062static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040064static const struct vm_operations_struct nfs_file_vm_ops;
Trond Myklebust94387fb2007-07-22 17:09:05 -040065
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080066const struct file_operations nfs_file_operations = {
Trond Myklebust980802e2005-06-13 11:14:01 -040067 .llseek = nfs_file_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 .read = do_sync_read,
69 .write = do_sync_write,
Badari Pulavarty027445c2006-09-30 23:28:46 -070070 .aio_read = nfs_file_read,
71 .aio_write = nfs_file_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 .mmap = nfs_file_mmap,
73 .open = nfs_file_open,
74 .flush = nfs_file_flush,
75 .release = nfs_file_release,
Chuck Lever54917782008-05-27 16:29:07 -040076 .fsync = nfs_file_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 .lock = nfs_lock,
78 .flock = nfs_flock,
Jens Axboef0930ff2007-06-01 11:51:43 +020079 .splice_read = nfs_file_splice_read,
Suresh Jayaramanbf40d342009-06-17 18:02:09 -070080 .splice_write = nfs_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 .check_flags = nfs_check_flags,
J. Bruce Fields370f6592007-06-08 15:23:34 -040082 .setlease = nfs_setlease,
Linus Torvalds1da177e2005-04-16 15:20:36 -070083};
84
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080085const struct inode_operations nfs_file_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 .permission = nfs_permission,
87 .getattr = nfs_getattr,
88 .setattr = nfs_setattr,
89};
90
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000091#ifdef CONFIG_NFS_V3
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080092const struct inode_operations nfs3_file_inode_operations = {
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000093 .permission = nfs_permission,
94 .getattr = nfs_getattr,
95 .setattr = nfs_setattr,
96 .listxattr = nfs3_listxattr,
97 .getxattr = nfs3_getxattr,
98 .setxattr = nfs3_setxattr,
99 .removexattr = nfs3_removexattr,
100};
101#endif /* CONFIG_NFS_v3 */
102
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103/* Hack for future NFS swap support */
104#ifndef IS_SWAPFILE
105# define IS_SWAPFILE(inode) (0)
106#endif
107
108static int nfs_check_flags(int flags)
109{
110 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
111 return -EINVAL;
112
113 return 0;
114}
115
116/*
117 * Open file
118 */
119static int
120nfs_file_open(struct inode *inode, struct file *filp)
121{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 int res;
123
Chuck Lever6da24bc2008-06-11 17:55:58 -0400124 dprintk("NFS: open file(%s/%s)\n",
Chuck Levercc0dd2d2008-06-11 17:55:42 -0400125 filp->f_path.dentry->d_parent->d_name.name,
126 filp->f_path.dentry->d_name.name);
127
Chuck Leverc2459dc2010-02-01 14:17:14 -0500128 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 res = nfs_check_flags(filp->f_flags);
130 if (res)
131 return res;
132
Trond Myklebust46cb6502008-06-11 16:32:46 -0400133 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 return res;
135}
136
137static int
138nfs_file_release(struct inode *inode, struct file *filp)
139{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400140 dprintk("NFS: release(%s/%s)\n",
Rakib Mullick6f276e42011-11-01 12:16:15 +0600141 filp->f_path.dentry->d_parent->d_name.name,
142 filp->f_path.dentry->d_name.name);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400143
Chuck Lever91d5b472006-03-20 13:44:14 -0500144 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
Trond Myklebust46cb6502008-06-11 16:32:46 -0400145 return nfs_release(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
147
Trond Myklebust980802e2005-06-13 11:14:01 -0400148/**
149 * nfs_revalidate_size - Revalidate the file size
150 * @inode - pointer to inode struct
151 * @file - pointer to struct file
152 *
153 * Revalidates the file length. This is basically a wrapper around
154 * nfs_revalidate_inode() that takes into account the fact that we may
155 * have cached writes (in which case we don't care about the server's
156 * idea of what the file length is), or O_DIRECT (in which case we
157 * shouldn't trust the cache).
158 */
159static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
160{
161 struct nfs_server *server = NFS_SERVER(inode);
162 struct nfs_inode *nfsi = NFS_I(inode);
163
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400164 if (nfs_have_delegated_attributes(inode))
165 goto out_noreval;
166
Trond Myklebust980802e2005-06-13 11:14:01 -0400167 if (filp->f_flags & O_DIRECT)
168 goto force_reval;
Trond Myklebustd7cf8dd2010-04-16 16:42:46 -0400169 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
170 goto force_reval;
171 if (nfs_attribute_timeout(inode))
172 goto force_reval;
173out_noreval:
174 return 0;
Trond Myklebust980802e2005-06-13 11:14:01 -0400175force_reval:
176 return __nfs_revalidate_inode(server, inode);
177}
178
179static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
180{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400181 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
Chuck Leverb84e06c2008-06-11 17:55:34 -0400182 filp->f_path.dentry->d_parent->d_name.name,
183 filp->f_path.dentry->d_name.name,
184 offset, origin);
185
Josef Bacik06222e42011-07-18 13:21:38 -0400186 /*
187 * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
188 * the cached file length
189 */
190 if (origin != SEEK_SET || origin != SEEK_CUR) {
Trond Myklebust980802e2005-06-13 11:14:01 -0400191 struct inode *inode = filp->f_mapping->host;
Trond Myklebustd5e66342008-09-23 17:28:35 -0400192
Trond Myklebust980802e2005-06-13 11:14:01 -0400193 int retval = nfs_revalidate_file_size(inode, filp);
194 if (retval < 0)
195 return (loff_t)retval;
Andi Kleen79835a712011-09-15 16:06:52 -0700196 }
Trond Myklebustd5e66342008-09-23 17:28:35 -0400197
Andi Kleen79835a712011-09-15 16:06:52 -0700198 return generic_file_llseek(filp, offset, origin);
Trond Myklebust980802e2005-06-13 11:14:01 -0400199}
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201/*
202 * Flush all dirty pages, and check for write errors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 */
204static int
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -0700205nfs_file_flush(struct file *file, fl_owner_t id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400207 struct dentry *dentry = file->f_path.dentry;
208 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Chuck Lever6da24bc2008-06-11 17:55:58 -0400210 dprintk("NFS: flush(%s/%s)\n",
211 dentry->d_parent->d_name.name,
212 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Chuck Leverc2459dc2010-02-01 14:17:14 -0500214 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if ((file->f_mode & FMODE_WRITE) == 0)
216 return 0;
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400217
Trond Myklebust7fe5c392009-03-19 15:35:50 -0400218 /* Flush writes to the server and return any errors */
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400219 return vfs_fsync(file, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220}
221
222static ssize_t
Badari Pulavarty027445c2006-09-30 23:28:46 -0700223nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
224 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800226 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 struct inode * inode = dentry->d_inode;
228 ssize_t result;
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 if (iocb->ki_filp->f_flags & O_DIRECT)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700231 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Chuck Lever6da24bc2008-06-11 17:55:58 -0400233 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 dentry->d_parent->d_name.name, dentry->d_name.name,
Rakib Mullick6f276e42011-11-01 12:16:15 +0600235 (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Trond Myklebust44b11872006-05-25 01:40:59 -0400237 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500238 if (!result) {
Badari Pulavarty027445c2006-09-30 23:28:46 -0700239 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
Chuck Lever4184dcf2010-02-01 14:17:23 -0500240 if (result > 0)
241 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 return result;
244}
245
246static ssize_t
Jens Axboef0930ff2007-06-01 11:51:43 +0200247nfs_file_splice_read(struct file *filp, loff_t *ppos,
248 struct pipe_inode_info *pipe, size_t count,
249 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800251 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 struct inode *inode = dentry->d_inode;
253 ssize_t res;
254
Chuck Lever6da24bc2008-06-11 17:55:58 -0400255 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 dentry->d_parent->d_name.name, dentry->d_name.name,
257 (unsigned long) count, (unsigned long long) *ppos);
258
Trond Myklebust44b11872006-05-25 01:40:59 -0400259 res = nfs_revalidate_mapping(inode, filp->f_mapping);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500260 if (!res) {
Jens Axboef0930ff2007-06-01 11:51:43 +0200261 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
Chuck Leveraa2f1ef2010-02-01 14:17:32 -0500262 if (res > 0)
263 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
264 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 return res;
266}
267
268static int
269nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
270{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800271 struct dentry *dentry = file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 struct inode *inode = dentry->d_inode;
273 int status;
274
Chuck Lever6da24bc2008-06-11 17:55:58 -0400275 dprintk("NFS: mmap(%s/%s)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 dentry->d_parent->d_name.name, dentry->d_name.name);
277
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400278 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
279 * so we call that before revalidating the mapping
280 */
281 status = generic_file_mmap(file, vma);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400282 if (!status) {
283 vma->vm_ops = &nfs_file_vm_ops;
Trond Myklebuste1ebfd32009-03-11 14:37:54 -0400284 status = nfs_revalidate_mapping(inode, file->f_mapping);
Trond Myklebust94387fb2007-07-22 17:09:05 -0400285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 return status;
287}
288
289/*
290 * Flush any dirty pages for this process, and check for write errors.
291 * The return status from this call provides a reliable indication of
292 * whether any write errors occurred for this process.
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400293 *
294 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
295 * disk, but it retrieves and clears ctx->error after synching, despite
296 * the two being set at the same time in nfs_context_set_write_error().
297 * This is because the former is used to notify the _next_ call to
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300298 * nfs_file_write() that a write error occurred, and hence cause it to
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400299 * fall back to doing a synchronous write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 */
301static int
Josef Bacik02c24a82011-07-16 20:44:56 -0400302nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200304 struct dentry *dentry = file->f_path.dentry;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400305 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 struct inode *inode = dentry->d_inode;
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400307 int have_error, status;
308 int ret = 0;
309
Chuck Lever6da24bc2008-06-11 17:55:58 -0400310 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
Chuck Lever54917782008-05-27 16:29:07 -0400311 dentry->d_parent->d_name.name, dentry->d_name.name,
312 datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
Josef Bacik02c24a82011-07-16 20:44:56 -0400314 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
315 if (ret)
316 return ret;
317 mutex_lock(&inode->i_mutex);
318
Chuck Lever91d5b472006-03-20 13:44:14 -0500319 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400320 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
321 status = nfs_commit_inode(inode, FLUSH_SYNC);
322 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
323 if (have_error)
324 ret = xchg(&ctx->error, 0);
J. R. Okajima07020992010-08-11 13:10:16 -0400325 if (!ret && status < 0)
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400326 ret = status;
Andy Adamson863a3c62011-03-23 13:27:54 +0000327 if (!ret && !datasync)
328 /* application has asked for meta-data sync */
Andy Adamsonef311532011-03-12 02:58:10 -0500329 ret = pnfs_layoutcommit_inode(inode, true);
Josef Bacik02c24a82011-07-16 20:44:56 -0400330 mutex_unlock(&inode->i_mutex);
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400331 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332}
333
334/*
Peter Staubach38c73042009-08-10 08:54:16 -0400335 * Decide whether a read/modify/write cycle may be more efficient
336 * then a modify/write/read cycle when writing to a page in the
337 * page cache.
338 *
339 * The modify/write/read cycle may occur if a page is read before
340 * being completely filled by the writer. In this situation, the
341 * page must be completely written to stable storage on the server
342 * before it can be refilled by reading in the page from the server.
343 * This can lead to expensive, small, FILE_SYNC mode writes being
344 * done.
345 *
346 * It may be more efficient to read the page first if the file is
347 * open for reading in addition to writing, the page is not marked
348 * as Uptodate, it is not dirty or waiting to be committed,
349 * indicating that it was previously allocated and then modified,
350 * that there were valid bytes of data in that range of the file,
351 * and that the new data won't completely replace the old data in
352 * that range of the file.
353 */
354static int nfs_want_read_modify_write(struct file *file, struct page *page,
355 loff_t pos, unsigned len)
356{
357 unsigned int pglen = nfs_page_length(page);
358 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
359 unsigned int end = offset + len;
360
361 if ((file->f_mode & FMODE_READ) && /* open for read? */
362 !PageUptodate(page) && /* Uptodate? */
363 !PagePrivate(page) && /* i/o request already? */
364 pglen && /* valid bytes of file? */
365 (end < pglen || offset)) /* replace all valid bytes? */
366 return 1;
367 return 0;
368}
369
370/*
Nick Piggin4899f9c2007-10-16 01:25:16 -0700371 * This does the "real" work of the write. We must allocate and lock the
372 * page to be sent back to the generic routine, which then copies the
373 * data from user space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 *
375 * If the writer ends up delaying the write, the writer needs to
376 * increment the page use counts until he is done with the page.
377 */
Nick Piggin4899f9c2007-10-16 01:25:16 -0700378static int nfs_write_begin(struct file *file, struct address_space *mapping,
379 loff_t pos, unsigned len, unsigned flags,
380 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700382 int ret;
Peter Staubach38c73042009-08-10 08:54:16 -0400383 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700384 struct page *page;
Peter Staubach38c73042009-08-10 08:54:16 -0400385 int once_thru = 0;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700386
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400387 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
388 file->f_path.dentry->d_parent->d_name.name,
389 file->f_path.dentry->d_name.name,
390 mapping->host->i_ino, len, (long long) pos);
391
Peter Staubach38c73042009-08-10 08:54:16 -0400392start:
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400393 /*
394 * Prevent starvation issues if someone is doing a consistency
395 * sync-to-disk
396 */
397 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
398 nfs_wait_bit_killable, TASK_KILLABLE);
399 if (ret)
400 return ret;
401
Nick Piggin54566b22009-01-04 12:00:53 -0800402 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700403 if (!page)
404 return -ENOMEM;
405 *pagep = page;
406
407 ret = nfs_flush_incompatible(file, page);
408 if (ret) {
409 unlock_page(page);
410 page_cache_release(page);
Peter Staubach38c73042009-08-10 08:54:16 -0400411 } else if (!once_thru &&
412 nfs_want_read_modify_write(file, page, pos, len)) {
413 once_thru = 1;
414 ret = nfs_readpage(file, page);
415 page_cache_release(page);
416 if (!ret)
417 goto start;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700418 }
419 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
Nick Piggin4899f9c2007-10-16 01:25:16 -0700422static int nfs_write_end(struct file *file, struct address_space *mapping,
423 loff_t pos, unsigned len, unsigned copied,
424 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
Nick Piggin4899f9c2007-10-16 01:25:16 -0700426 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
427 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400429 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
430 file->f_path.dentry->d_parent->d_name.name,
431 file->f_path.dentry->d_name.name,
432 mapping->host->i_ino, len, (long long) pos);
433
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400434 /*
435 * Zero any uninitialised parts of the page, and then mark the page
436 * as up to date if it turns out that we're extending the file.
437 */
438 if (!PageUptodate(page)) {
439 unsigned pglen = nfs_page_length(page);
440 unsigned end = offset + len;
441
442 if (pglen == 0) {
443 zero_user_segments(page, 0, offset,
444 end, PAGE_CACHE_SIZE);
445 SetPageUptodate(page);
446 } else if (end >= pglen) {
447 zero_user_segment(page, end, PAGE_CACHE_SIZE);
448 if (offset == 0)
449 SetPageUptodate(page);
450 } else
451 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
452 }
453
Nick Piggin4899f9c2007-10-16 01:25:16 -0700454 status = nfs_updatepage(file, page, offset, copied);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700455
456 unlock_page(page);
457 page_cache_release(page);
458
Chuck Lever3d509e52007-12-20 14:55:04 -0500459 if (status < 0)
460 return status;
461 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
463
David Howells6b9b3512009-04-03 16:42:41 +0100464/*
465 * Partially or wholly invalidate a page
466 * - Release the private state associated with a page if undergoing complete
467 * page invalidation
David Howells545db452009-04-03 16:42:44 +0100468 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100469 * - Caller holds page lock
470 */
NeilBrown2ff28e22006-03-26 01:37:18 -0800471static void nfs_invalidate_page(struct page *page, unsigned long offset)
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500472{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400473 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
474
Trond Myklebust1c759502006-10-09 16:18:38 -0400475 if (offset != 0)
476 return;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400477 /* Cancel any unstarted writes on this page */
Trond Myklebust1b3b4a12007-08-28 10:29:36 -0400478 nfs_wb_page_cancel(page->mapping->host, page);
David Howells545db452009-04-03 16:42:44 +0100479
480 nfs_fscache_invalidate_page(page, page->mapping->host);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500481}
482
David Howells6b9b3512009-04-03 16:42:41 +0100483/*
484 * Attempt to release the private state associated with a page
David Howells545db452009-04-03 16:42:44 +0100485 * - Called if either PG_private or PG_fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100486 * - Caller holds page lock
487 * - Return true (may release page) or false (may not)
488 */
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500489static int nfs_release_page(struct page *page, gfp_t gfp)
490{
Trond Myklebustb608b282010-07-30 15:31:54 -0400491 struct address_space *mapping = page->mapping;
492
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400493 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
494
Trond Myklebustd812e572010-03-19 13:55:17 -0400495 /* Only do I/O if gfp is a superset of GFP_KERNEL */
Trond Myklebustb608b282010-07-30 15:31:54 -0400496 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL) {
497 int how = FLUSH_SYNC;
498
499 /* Don't let kswapd deadlock waiting for OOM RPC calls */
500 if (current_is_kswapd())
501 how = 0;
502 nfs_commit_inode(mapping->host, how);
503 }
Trond Myklebuste3db7692007-01-10 23:15:39 -0800504 /* If PagePrivate() is set, then the page is not freeable */
David Howells545db452009-04-03 16:42:44 +0100505 if (PagePrivate(page))
506 return 0;
507 return nfs_fscache_release_page(page, gfp);
Trond Myklebuste3db7692007-01-10 23:15:39 -0800508}
509
David Howells6b9b3512009-04-03 16:42:41 +0100510/*
511 * Attempt to clear the private state associated with a page when an error
512 * occurs that requires the cached contents of an inode to be written back or
513 * destroyed
David Howells545db452009-04-03 16:42:44 +0100514 * - Called if either PG_private or fscache is set on the page
David Howells6b9b3512009-04-03 16:42:41 +0100515 * - Caller holds page lock
516 * - Return 0 if successful, -error otherwise
517 */
Trond Myklebuste3db7692007-01-10 23:15:39 -0800518static int nfs_launder_page(struct page *page)
519{
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400520 struct inode *inode = page->mapping->host;
David Howells545db452009-04-03 16:42:44 +0100521 struct nfs_inode *nfsi = NFS_I(inode);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400522
523 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
524 inode->i_ino, (long long)page_offset(page));
525
David Howells545db452009-04-03 16:42:44 +0100526 nfs_fscache_wait_on_page_write(nfsi, page);
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400527 return nfs_wb_page(inode, page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500528}
529
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700530const struct address_space_operations nfs_file_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 .readpage = nfs_readpage,
532 .readpages = nfs_readpages,
Trond Myklebust9cccef92007-07-22 17:09:05 -0400533 .set_page_dirty = __set_page_dirty_nobuffers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 .writepage = nfs_writepage,
535 .writepages = nfs_writepages,
Nick Piggin4899f9c2007-10-16 01:25:16 -0700536 .write_begin = nfs_write_begin,
537 .write_end = nfs_write_end,
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500538 .invalidatepage = nfs_invalidate_page,
539 .releasepage = nfs_release_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 .direct_IO = nfs_direct_IO,
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400541 .migratepage = nfs_migrate_page,
Trond Myklebuste3db7692007-01-10 23:15:39 -0800542 .launder_page = nfs_launder_page,
Andi Kleenf590f332009-09-16 11:50:17 +0200543 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544};
545
David Howells6b9b3512009-04-03 16:42:41 +0100546/*
547 * Notification that a PTE pointing to an NFS page is about to be made
548 * writable, implying that someone is about to modify the page through a
549 * shared-writable mapping
550 */
Nick Pigginc2ec1752009-03-31 15:23:21 -0700551static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Trond Myklebust94387fb2007-07-22 17:09:05 -0400552{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700553 struct page *page = vmf->page;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400554 struct file *filp = vma->vm_file;
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400555 struct dentry *dentry = filp->f_path.dentry;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400556 unsigned pagelen;
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400557 int ret = VM_FAULT_NOPAGE;
Nick Piggin4899f9c2007-10-16 01:25:16 -0700558 struct address_space *mapping;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400559
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400560 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
561 dentry->d_parent->d_name.name, dentry->d_name.name,
562 filp->f_mapping->host->i_ino,
563 (long long)page_offset(page));
564
David Howells545db452009-04-03 16:42:44 +0100565 /* make sure the cache has finished storing the page */
566 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
567
Trond Myklebust94387fb2007-07-22 17:09:05 -0400568 lock_page(page);
Nick Piggin4899f9c2007-10-16 01:25:16 -0700569 mapping = page->mapping;
Chuck Leverb7eaefa2008-06-11 17:55:50 -0400570 if (mapping != dentry->d_inode->i_mapping)
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500571 goto out_unlock;
572
Trond Myklebust94387fb2007-07-22 17:09:05 -0400573 pagelen = nfs_page_length(page);
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500574 if (pagelen == 0)
575 goto out_unlock;
576
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400577 ret = VM_FAULT_LOCKED;
578 if (nfs_flush_incompatible(filp, page) == 0 &&
579 nfs_updatepage(filp, page, 0, pagelen) == 0)
580 goto out;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500581
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400582 ret = VM_FAULT_SIGBUS;
Trond Myklebust8b1f9ee2008-01-22 17:13:06 -0500583out_unlock:
Trond Myklebust94387fb2007-07-22 17:09:05 -0400584 unlock_page(page);
Trond Myklebustbc4866b2010-10-04 17:59:08 -0400585out:
586 return ret;
Trond Myklebust94387fb2007-07-22 17:09:05 -0400587}
588
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400589static const struct vm_operations_struct nfs_file_vm_ops = {
Trond Myklebust94387fb2007-07-22 17:09:05 -0400590 .fault = filemap_fault,
591 .page_mkwrite = nfs_vm_page_mkwrite,
592};
593
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400594static int nfs_need_sync_write(struct file *filp, struct inode *inode)
595{
596 struct nfs_open_context *ctx;
597
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100598 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400599 return 1;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400600 ctx = nfs_file_open_context(filp);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400601 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
602 return 1;
603 return 0;
604}
605
Badari Pulavarty027445c2006-09-30 23:28:46 -0700606static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
607 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800609 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 struct inode * inode = dentry->d_inode;
Chuck Lever7e381172010-02-01 14:17:41 -0500611 unsigned long written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 ssize_t result;
Badari Pulavarty027445c2006-09-30 23:28:46 -0700613 size_t count = iov_length(iov, nr_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (iocb->ki_filp->f_flags & O_DIRECT)
Badari Pulavarty027445c2006-09-30 23:28:46 -0700616 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
Chuck Lever6da24bc2008-06-11 17:55:58 -0400618 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 dentry->d_parent->d_name.name, dentry->d_name.name,
Chuck Lever6da24bc2008-06-11 17:55:58 -0400620 (unsigned long) count, (long long) pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
622 result = -EBUSY;
623 if (IS_SWAPFILE(inode))
624 goto out_swapfile;
Trond Myklebust7d52e862005-06-22 17:16:30 +0000625 /*
626 * O_APPEND implies that we must revalidate the file length.
627 */
628 if (iocb->ki_filp->f_flags & O_APPEND) {
629 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
630 if (result)
631 goto out;
Trond Myklebustfe51beec2005-06-22 17:16:30 +0000632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
634 result = count;
635 if (!count)
636 goto out;
637
Badari Pulavarty027445c2006-09-30 23:28:46 -0700638 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
Chuck Lever7e381172010-02-01 14:17:41 -0500639 if (result > 0)
640 written = result;
641
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100642 /* Return error values for O_DSYNC and IS_SYNC() */
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400643 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400644 int err = vfs_fsync(iocb->ki_filp, 0);
Trond Myklebust200baa22006-12-05 00:35:40 -0500645 if (err < 0)
646 result = err;
647 }
Chuck Lever7e381172010-02-01 14:17:41 -0500648 if (result > 0)
649 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650out:
651 return result;
652
653out_swapfile:
654 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
655 goto out;
656}
657
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700658static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
659 struct file *filp, loff_t *ppos,
660 size_t count, unsigned int flags)
661{
662 struct dentry *dentry = filp->f_path.dentry;
663 struct inode *inode = dentry->d_inode;
Chuck Lever7e381172010-02-01 14:17:41 -0500664 unsigned long written = 0;
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700665 ssize_t ret;
666
667 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
668 dentry->d_parent->d_name.name, dentry->d_name.name,
669 (unsigned long) count, (unsigned long long) *ppos);
670
671 /*
672 * The combination of splice and an O_APPEND destination is disallowed.
673 */
674
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700675 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
Chuck Lever7e381172010-02-01 14:17:41 -0500676 if (ret > 0)
677 written = ret;
678
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700679 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
Trond Myklebustaf7fa162010-07-31 14:29:06 -0400680 int err = vfs_fsync(filp, 0);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700681 if (err < 0)
682 ret = err;
683 }
Chuck Lever7e381172010-02-01 14:17:41 -0500684 if (ret > 0)
685 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
Suresh Jayaramanbf40d342009-06-17 18:02:09 -0700686 return ret;
687}
688
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400689static int
690do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
692 struct inode *inode = filp->f_mapping->host;
693 int status = 0;
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000694 unsigned int saved_type = fl->fl_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Trond Myklebust039c4d72005-10-18 14:20:16 -0700696 /* Try local locking first */
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400697 posix_test_lock(filp, fl);
698 if (fl->fl_type != F_UNLCK) {
699 /* found a conflict */
Trond Myklebust039c4d72005-10-18 14:20:16 -0700700 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 }
Sergey Vlasov21ac19d2010-11-28 21:04:05 +0000702 fl->fl_type = saved_type;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700703
704 if (nfs_have_delegation(inode, FMODE_READ))
705 goto out_noconflict;
706
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400707 if (is_local)
Trond Myklebust039c4d72005-10-18 14:20:16 -0700708 goto out_noconflict;
709
710 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
711out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 return status;
Trond Myklebust039c4d72005-10-18 14:20:16 -0700713out_noconflict:
714 fl->fl_type = F_UNLCK;
715 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716}
717
718static int do_vfs_lock(struct file *file, struct file_lock *fl)
719{
720 int res = 0;
721 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
722 case FL_POSIX:
723 res = posix_lock_file_wait(file, fl);
724 break;
725 case FL_FLOCK:
726 res = flock_lock_file_wait(file, fl);
727 break;
728 default:
729 BUG();
730 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 return res;
732}
733
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400734static int
735do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736{
737 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 int status;
739
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 /*
741 * Flush all pending writes before doing anything
742 * with locks..
743 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500744 nfs_sync_mapping(filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
746 /* NOTE: special case
747 * If we're signalled while cleaning up locks on process exit, we
748 * still need to complete the unlock.
749 */
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400750 /*
751 * Use local locking if mounted with "-onolock" or with appropriate
752 * "-olocal_lock="
753 */
754 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
756 else
757 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return status;
759}
760
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400761static int
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700762is_time_granular(struct timespec *ts) {
763 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
764}
765
766static int
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400767do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
769 struct inode *inode = filp->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 int status;
771
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 /*
773 * Flush all pending writes before doing anything
774 * with locks..
775 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500776 status = nfs_sync_mapping(filp->f_mapping);
777 if (status != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 goto out;
779
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400780 /*
781 * Use local locking if mounted with "-onolock" or with appropriate
782 * "-olocal_lock="
783 */
784 if (!is_local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
Trond Myklebustc4d7c402008-04-01 20:26:52 -0400786 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 status = do_vfs_lock(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 if (status < 0)
789 goto out;
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 /*
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700792 * Revalidate the cache if the server has time stamps granular
793 * enough to detect subsecond changes. Otherwise, clear the
794 * cache to prevent missing any changes.
795 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 * This makes locking act as a cache coherency point.
797 */
Trond Myklebust29884df2005-12-13 16:13:54 -0500798 nfs_sync_mapping(filp->f_mapping);
Ricardo Labiaga6b967242010-10-12 16:30:05 -0700799 if (!nfs_have_delegation(inode, FMODE_READ)) {
800 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
801 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
802 else
803 nfs_zap_caches(inode);
804 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 return status;
807}
808
809/*
810 * Lock a (portion of) a file
811 */
812static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
813{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400814 struct inode *inode = filp->f_mapping->host;
Trond Myklebust21162712008-05-20 19:34:39 -0400815 int ret = -ENOLCK;
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400816 int is_local = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Chuck Lever6da24bc2008-06-11 17:55:58 -0400818 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
819 filp->f_path.dentry->d_parent->d_name.name,
820 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 fl->fl_type, fl->fl_flags,
822 (long long)fl->fl_start, (long long)fl->fl_end);
Chuck Lever6da24bc2008-06-11 17:55:58 -0400823
Chuck Lever91d5b472006-03-20 13:44:14 -0500824 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826 /* No mandatory locks over NFS */
Pavel Emelyanovdfad9442007-10-01 14:41:15 -0700827 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
Trond Myklebust21162712008-05-20 19:34:39 -0400828 goto out_err;
829
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400830 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
831 is_local = 1;
832
Trond Myklebust21162712008-05-20 19:34:39 -0400833 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
834 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
835 if (ret < 0)
836 goto out_err;
837 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
839 if (IS_GETLK(cmd))
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400840 ret = do_getlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400841 else if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400842 ret = do_unlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400843 else
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400844 ret = do_setlk(filp, cmd, fl, is_local);
Trond Myklebust21162712008-05-20 19:34:39 -0400845out_err:
846 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847}
848
849/*
850 * Lock a (portion of) a file
851 */
852static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
853{
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400854 struct inode *inode = filp->f_mapping->host;
855 int is_local = 0;
856
Chuck Lever6da24bc2008-06-11 17:55:58 -0400857 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
858 filp->f_path.dentry->d_parent->d_name.name,
859 filp->f_path.dentry->d_name.name,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 fl->fl_type, fl->fl_flags);
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 if (!(fl->fl_flags & FL_FLOCK))
863 return -ENOLCK;
864
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400865 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
866 is_local = 1;
867
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 /* We're simulating flock() locks using posix locks on the server */
869 fl->fl_owner = (fl_owner_t)filp;
870 fl->fl_start = 0;
871 fl->fl_end = OFFSET_MAX;
872
873 if (fl->fl_type == F_UNLCK)
Suresh Jayaraman5eebde22010-09-23 08:55:58 -0400874 return do_unlk(filp, cmd, fl, is_local);
875 return do_setlk(filp, cmd, fl, is_local);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876}
J. Bruce Fields370f6592007-06-08 15:23:34 -0400877
Chuck Lever6da24bc2008-06-11 17:55:58 -0400878/*
879 * There is no protocol support for leases, so we have no way to implement
880 * them correctly in the face of opens by other clients.
881 */
J. Bruce Fields370f6592007-06-08 15:23:34 -0400882static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
883{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400884 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
885 file->f_path.dentry->d_parent->d_name.name,
886 file->f_path.dentry->d_name.name, arg);
J. Bruce Fields370f6592007-06-08 15:23:34 -0400887 return -EINVAL;
888}